Jan 31, 2025 · 51m · saastr
The Future of AI in B2B SaaS: Insights from Synthesia and Theory Ventures. Hosted by Jason Lemkin
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
At SaaStr Digital AI Day 2025, host Jason Lemkin joins Synthesia CEO Victor Ripabelli and Theory Ventures founder Tomasz Tunguz to analyze shifting AI unit economics, enterprise monetization debates, and defensible B2B platform strategies.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Jason holds 47.5% of the talking time here. How this is scored →
Jason as informed peer 6.1 Guest teaching 4.3 Guest disagreement 2.3 Jason pushing back 4.0
speaking balance: gold is Jason, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 40:04 Victor dismisses the agent buzzword obsession
Victor rejects the industry's framing of AI agents, passionately arguing that calling LLM-driven features 'coworkers' is absurd marketing rather than a meaningful technical distinction.
Hardest push from Jason ▶ 16:16 Jason challenges outcome-based AI monetizationJason refuses the prevailing VC narrative that enterprise software can hike pricing 50-70% for AI add-ons, labeling it an unsustainable fantasy that ignores deflationary customer expectations.
Biggest teaching moment ▶ 12:05 Tomasz breaks down DeepSeek's gross margin impactTomasz clearly educates the audience on AI SaaS economics, showing how a 95% reduction in inference costs shifts application gross margins from 0-30% back to classic 75% software margins.
Jason holds their own ▶ 20:12 Jason refutes seat-death narrative with historical SaaS metricsJason cites specific KeyBank SaaS industry benchmark studies to prove seat pricing was never the majority of SaaS revenue, countering shallow social media hot takes with empirical data.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Jason as informed peer | Guest teaching | Guest disagreement | Jason pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome to SaaStr Digital AI Day 2025 and Guest Introductions | 5 | 3 | 1 | 2 | Jason sets up the session with context on venture fund growth and Synthesia's market positioning. Victor explains how Synthesia collapsed the fragmented video value chain into a single enterprise platform. | |
| Escaping the Uncanny Valley: Upgraded Models and Synthesia's TAM Expansion | 5 | 4 | 1 | 2 | Victor explains how starting in the unsexy L&D training niche provided a beachhead before model upgrades expanded their TAM into marketing. Jason connects improved avatar fidelity to escaping the uncanny valley and unlocking market pull. | |
| Analyzing the DeepSeek Disruption: Benchmark Hype vs. Enterprise Workflows | 5 | 5 | 2 | 2 | Tomasz breaks down how DeepSeek slashes inference costs by 95% to restore traditional SaaS gross margins. Victor contextualizes the hype, noting that enterprise customers prioritize workflows over raw benchmark scores. | |
| AI Monetization, Usage Deflation, and Shifting from Seats to Outcomes | 7 | 4 | 3 | 6 | Jason challenges the idea that SaaS vendors can charge steep premiums for AI add-ons, citing Gorgias and HubSpot's free agent rollouts. Tomasz pushes back that software pricing has never been strictly cost-plus and lower inference enables expanded usage. | |
| Seat-Based vs. Outcome Pricing: Enterprise Buyer Preferences and Budget Certainty | 7 | 3 | 2 | 5 | Jason cites historical KeyBank SaaS data to debunk claims that per-seat pricing is dead, emphasizing that enterprise procurement requires budget certainty. Victor confirms Synthesia still sells primarily via seats for core creation tools. | |
| The AI Venture Landscape: Flight to Quality, Wrapper Fatigue, and Durable Revenue | 6 | 5 | 2 | 3 | Jason presents venture funding data showing concentrated capital in fewer growth deals. Tomasz compares API wrappers to the Groupon era, while Victor explains how enterprise renewals separate durable business value from experimental demo budgets. | |
| Agent Proliferation, Vendor Fatigue, and the Reality of Enterprise Stack Expansion | 7 | 6 | 5 | 7 | Jason uses Zylo stack data to challenge the notion that CIOs will manage hundreds of discrete agent vendors. Victor forcefully dismisses agent terminology altogether, calling it anthropomorphized marketing hype for software features. | |
| Engineering Productivity, AI Augmentation, and Concluding Remarks | 7 | 4 | 2 | 5 | Jason compares current VC mandates for lean AI-assisted engineering teams to 2000s offshoring mandates. Tomasz and Victor emphasize that AI augments humans rather than entirely eliminating organizational scaling requirements in enterprise SaaS. |
Statements from this episode (26)
Tunguz: Theory Ventures raised closer to $700M across two funds in 14 months
“A little less than that. Closer to 700, but.”
Tunguz: Lead VC check sizes are up 42% in 18 months
“Right, but check sizes are up, the lead check sizes are up, 42% in 18 months.”
Ripabelli: No one has ever built a really big enterprise video company
“If you look at the video value chain, no one has ever managed to build a really big company about enterprise video.”
Ripabelli: Synthesia will launch new AI video models within six months
“So we're launching our new AI video models within the next three to six months, and they're a very massive upgrade, both in terms of quality and fidelity of the avatar.”
Ripabelli: Product marketing and customer support are Synthesia's largest segments
“So the last 12 to 18 months, we've seen a massive shift towards product marketing with customer support, which collectively those two are now actually the biggest segments that we serve.”
Tunguz: AI app gross margins will expand as inference costs drop 95%
“And when we were looking like a typical software companies, probably at 75% gross margin. Yeah. AI software application companies are like a zero to 30%, and now their single biggest cost has been reduced by 95% overnight, and they didn't have to do a thing.”
Tunguz predicts AI startup margins and multiples will match classic SaaS
“Before they used to have Challenging gross margins. And now they'll probably look the same as SaaS companies, which means you have pretty significant multiple and valuation expansion.”
Ripabelli: 99.9% of OpenAI and Anthropic users will not stay on DeepSeek
“But I think in two months, I think, 99.9% of the people who are using Anthropic and OpenAI today is going to go back to using their products, not DeepSeek's model.”
Ripabelli: Tech analysts talk benchmarks while real customers talk workflows
“I tweeted the other day that smart people, Talk about models and benchmarks, and customers talk about app flows, apps and workflows, right?”
Lemkin: Gorgias nears $100M ARR after optimizing $1M per month AI cost
“One I invested in the contact center, which is all about AI agents called gorgeous free commerce. They're at almost a hundred million. Okay. At first, the AI was a million bucks a month, right? To deploy these agents. Now it's fine. They've optimized it.”
Lemkin: HubSpot is currently not charging customers for AI agents
“When I talked with Brian Halligan the other day, I didn't realize Dharmesh right now HubSpot isn't charging for their agents.”
Tunguz: 95% cheaper AI models will drive 10x-100x usage increases
“If DeepSeq like models with great reasoning, and 95% cheaper. Not for companies will just like 10 X or a hundred X the use and probably keep the margins the same to get to a much better user experience. As a way of differentiation, right?”
Lemkin: Per-seat pricing was under 50% of SaaS revenue historically
“And it turns out going back to the beginning per seat was less than 50% of the SaaS revenue. When you add in API pricing, other pricing, just because we think everyone talks about blah, blah, blah, Salesforce, it doesn't mean it's most of software. And so it's…”
Ripabelli: Synthesia still predominantly sells seat-based licenses
“We still sell seats predominantly. I think it depends just a lot of the product, right? I think the way to think about Sulthesia right now is PowerPoint two point zero, right? And like PowerPoint is one of those things where we have thousands of people using t…”
Tunguz: Series B through D VC funding market is starting to recover
“After 2022, the series B through the D market was effectively dead and it's starting to come back.”
Tunguz: Software startups tripling growth without AI are a hard sell
“If you're tripling, doubling without AI, it's a hard sell.”
Tunguz: AI is in a "Groupon era" where GTM outpaces product differentiation
“It's a Groupon era for AI where one company takes off and then there are 15 to 20 others because you're wrapping the API, you have a lot of capability out of the gate, and then long-term product differentiation probably matters less than go-to-market execution…”
Ripabelli: Enterprise AI spending is driven by fear, not clear buyer need
“There's this weird thing in AI right now where it's very fear driven, and people are very willing to part ways with their money, both on consumer side, but also businesses to try out new things, right? Because everyone is scared that their business is going to…”
Lemkin: Zylo data shows zero net apps added to enterprise stacks
“There's Salesforce charging two bucks, but this was, there was one on the bottom right is the latest data from Zylo and they're all fairly consistent, which is basically over the last year, people didn't add any net apps. They didn't add any net apps to their …”
Tunguz: Enterprise app counts will flatline while total software spend rises
“And so I think maybe what we see is application count stays flat or maybe even decreases, but the total amount of software spending increases because it's capturing some fraction of that labor spend.”
Ripabelli: In five years, AI agent anthropomorphism will look ridiculous
“I think five years time, like, I think everyone's like, why the fuck did we spend five years talking about algorithms as being like a coworker? It's just like software that does things for us. And sometimes maybe you want to chat with it. That's great. That's …”
Ripabelli: Synthesia will launch an AI agent strategy just for the hype
“And that said, you know, we're probably launching an AI agent strategy soon because everybody likes to talk about this stuff, but nobody wants to talk about features.”
Tunguz: Portfolio startup keeping engineering headcount flat using AI IDEs
“We have one company growing really fast. Engineering headcount year over year will be flat. Because they deploy some AI enabled ID and they can write a lot more software.”
Ripabelli: Building a $100B Company Still Requires Massive Headcount
“I still think if you want to build a hundred billion dollar company, You've got to need a lot of salespeople. You've got to need a lot of engineers.”
Ripabelli: Fully Autonomous AI Agents Remain a Pipe Dream for Now
“Where AI really shines is still augmenting humans, right? In the case of us, we augment people to make videos. Cursor works so well because it puts the human in the driving seat and just makes you more efficient. I think this, like, pipe dream of agents, like,…”
Lemkin: S-tier AI agents will quickly replace B-tier and C-tier employees
“My limited experience is that an S-tier agent is better than a C-tier employee. This is what across the contact center is. You usually see that the agents are like in the, they can't do everything, but they're in the top 10% of CSAT because what they do, they …”